1/19 Structural versus Behavioral Remedies in the Deregulation of Electricity Markets: An Experimental Investigation Guided by Theory and Policy Concerns

Silvester van Koten, 1 Loyola de Palacio Chair, RSCAS, European University Institute, Florence CERGE-EI, Prague Silvester.VanKoten@eui.eu, slvstr@gmail.com

Andreas Ortmann, Australian School of Business, University of New South Wales, Sydney a.ortmann@unsw.edu.au, aortmann@gmail.com

Keywords: European Electricity Markets, Economics Experiments, Remedies, Forward Markets, Competition Policy

Abstract We experimentally study the effects of introducing a forward market and of increasing the number of competitors in a quantity-setting market. Our key interest was to better understand which of these two remedies enhances competition. Allaz and Vila (1993) theoretically showed that forward markets can have a pro-competitive effect. Le Coq and Orzen (2006) and Brandts, Pezanis-Christou and Schram (2008) experimentally investigated similar issues. All three experiments (including ours) support the prediction by Allaz and Vila (1993) – that introducing a forward market does indeed intensify competition. The results of the present study, however, differ from previous experimental results. In our experiment, the forward market is a more effective remedy than increasing the number of competitors. We argue that our results differ from Brandts et al.’s (2008) because they increased the number of competitors by entry, which thus increased the asset base and made production cheaper. In contrast, we increased the number of competitors by divestiture, which left the asset base constant. Our results address an important policy issue and provide tentative evidence on the competition-enhancing effect of forward markets, which are considered a behavioral remedy and are favored by the European Commission.

1

Corresponding author.

structural remedies and behavioral remedies. This gives scope for exercising market power” (European Commission 2007a. Indeed this is why Harvey and Hogan (2000) and Kamat and Oren (2004) question the competition-enhancing effect of forward markets. and Yurtoglu (2011.4 In line with Allaz and Vila’s (1993) predictions. however. 980) characterize behavioral remedies as measures that “… effectively tackle the market power concerns potentially raised by mergers without destroying efficiency enhancing synergies.. Indeed while increasing the number of competitors is a suitable benchmark for testing the competition-enhancing properties of forward markets. Introduction Mitigating market power is a central problem faced by policy makers and regulators.2 such as divestiture (structural) and forward markets (behavioral).” 3 Economics experiments are a form of empirical investigation that allows controlled testing of theoretically predicted effects (Rassenti. appears weak relative to increasing the number of competitors in the market. 8) “a measure that obliges the concerned undertaking(s) to act in a specific way. Allaz and Vila (1993) assume that competition is a one-shot game.. and generally maintain the high level of concentration of the pre-liberalization period. p. we note that the relative weakness of the forward market instrument has only been shown The European Commission (2006b. 2006a. p.11). they find that introducing forward markets intensifies competition. 4 Another paper addressing forward markets is Ferreira.7). Pezanis-Christou. the European Commission claims that EU electricity markets “… remain national in scope. p. Le Coq and Orzen (2006) and Brandts.2/19 1. but we comment on it in footnote 33 (results section). The most obvious one is the divestiture of an existing business. and Schram (2008) experimentally3 address the competitionenhancing effect of forward markets. The paper is currently unpublished. 2011). For example. low-cost robustness tests of various design and implementation characteristics. They allow. stating: “Structural remedies should only be imposed either where there is no equally effective behavioural remedy or where any equally effective behavioural remedy would be more burdensome for the undertaking concerned than the structural remedy” (European Commission. differ from Allaz and Vila’s (1993) theoretical portrayal. Allaz and Vila (1993) theoretically show that forward markets can mitigate market power by enhancing competition. The European Commission prefers behavioral remedies to structural ones. Gugler. Smith and Wilson 2002. Roth 2002). Therefore behavioral remedies may actually be less effective than structural remedies (Duso et al. Wolak (2001) and Van Eijkel and Moraga-Gonzalez (2010) find that forward trading may have increased aggregate supply in the Australian electricity market and the Dutch gas market. and of scenarios that are counterfactual to what currently exists. however. respectively. in particular. Real forward markets may.” Duso. This effect. 2 . 6) defines structural remedies as “changes to the structure of an undertaking. Kujal and Rassenti (2009). The Commission uses two broad types of remedies to address market power. For example. firms are likely to interact repeatedly in the real world. Empirical evidence on the competition-enhancing effect of forward markets is scant. p. however.” and behavioral remedies as (p.

6 See the Appendix for a mathematical exposition of this point in a market with Cournot competition.. as it unduly handicaps the competition) enhancing effects of forward markets (which do not affect the asset base). Positive and convex costs imply producers must make careful decisions. 2006). It is also interesting empirically. has no asset effect.e. (2008). In most industries. Moreover. For example. The two dominant electricity generators. as it has no asset effect. We argue that a suitable benchmark to study the effect of forward markets in capital-intensive industries is to increase of the number of competitors by divestiture rather than by entry. as it has played an important role in increasing competition in several markets. marginal costs are flat but sharply increase when capacity constraints become binding (e. introducing a forward market or a new competitor through divestiture requires no investment in new production assets. Denmark. the UK addressed market power in the late nineties by coaxing dominant electricity generators to divest plants. Divestiture. it makes no difference in the predictions if the number of competitors is increased by entry or divestiture. Specifically. in contrast. Increasing competition by divestiture is interesting theoretically and experimentally. Indeed. rather than by divestiture. building new power plants in the electricity industry is very costly: When competition is lacking but there is no shortage of electricity production capacity. as they can incur considerable losses if they produce too much.3/19 experimentally for industries with zero production cost (Le Coq and Orzen 2006) and for entry by new competitors (Brandts et al. 2002). (2008). and the As Le Coq and Orzen (2006) assume zero production cost.6 The resultant increase in production is thus a combined effect of more competitors (the competition-enhancing effect) and production assets (the asset effect).g. many describe the marginal cost of producing in the electricity generation industry as “hockey-stick” shaped. as in Brandts et al. which requires costly investment in new production assets. For example. NationalPower and PowerGen. France. The results of Le Coq and Orzen (2006) therefore cannot be generalized to industries with positive and convex costs. short-term costs are positive and сonvex due to a limited stock of (expensive) production assets. Newbery. i. which allows us to cleanly compare the competition-enhancing effect of introducing a forward market to the competition-enhancing effect of adding more competitors through divestiture.7 The benchmark is therefore biased. which resulted in lowered concentration (Green. We explain this in more detail in the Appendix. Belgium.. which means any given level of quantity is cheaper to produce due to the convex production function (the asset effect). introducing a new competitor through entry increases the total set of assets in the industry. 5 . Italy. entrance leads to wasteful duplication of assets (Green 1996).5 Unlike introducing a new competitor through entry. 7 While the increasing production assets may have positive effects. altogether divested 6GW in 1996 and another 8GW in 1999. the costs of creating these assets can be considerable.

4) competitors as M3 (M2. we also include a market with 2 competitors as a better approximation. each producer divests 1 ( n + 1) of their assets. these divested assets form a new independent producer. we report the results focusing on aggregate quantity. 10 See Van Koten and Ortmann (2011). p. 9 Quadratic marginal costs are considered a reasonable approximation to the marginal costs of electricity generators (Green and Newbery 1992. Borenstein. and underlying working hypotheses). 2006). Since the new EU member states have more concentrated markets. 8 . In section 5 we conclude. This divestiture of generation capacity is called virtual as no production capacity changes hand.11 When one competitor is added to a market with n producers by symmetric divestiture. (2008) but not to those with the M4 market as we created M4 by divestiture rather than entry. we drew (to the extent possible) on Brandts et al. we summarize the experimental procedures. and Van Koten and Ortmann (2008) for an overview of the EU member states and an explanation of how these numbers were determined. and production efficiency. (2008) includes an asset effect and thus provides a biased benchmark. M4 (M2) is created from M3 by symmetric divestiture (reverse symmetric divestiture). adding competitors by entry as in Brandts et al. (2008) and on Le Coq and Orzen (2006) to design and parameterize our experiment.4/19 Netherlands are using. and convex cost function for a firm in a market with 3 competitors: c3 [ q] = ∑ x =1 2 x 2 = (2 3)q 3 + q 2 + (1 3)q as Brandts et al. (2008).10 We refer to a market with 3 (2. 2006). efficiency. 11 The treatments with the M3 and M3F markets are thus identical to the ones in Brandts et al. In section 4. (2008) and Le Coq and Orzen (2006). we use the same demand function: p[Q ] = Max[0. In contrast to Brandts et al.. (2008). he sells part of his production capacity to other generators. In the following section we first discuss our experimental design (i. and the selling generator remains the owner of all its generation plants (Willems.e. For comparability. 2. To repeat. Experimental design As in Brandts et al. 2000 − 27Q ] for Q ≥ 0 . the basic parameterization. we use Cournot competition with a linear demand function. 5. Bushnell and Wolak 2002). and to a market with 3 (2) competitors and a forward market as M3F (M2F). M4). We compared the competition-enhancing effect of a forward market to the benchmark of increasing the number of competitors by divestiture. We thus eliminated the asset effect confound and isolated the competition effect. treatments. The q ) When a generator sells a Virtual Power Plant. or have used the auctioning of Virtual Power Plants8 to lower market power (Willems.9 A market with 3 competitors is a reasonable approximation for the old EU member states. In section 3. To make the comparison meaningful.

some of the rounded total costs are different. M3(F) and M4. but the discrepancy is negligible. Table 1 summarizes our treatments and indicates how they compare with Brandts et al. footnote 20. (2008). (2008). as the market has been created from the market with 3 producers by divestiture.5/19 new market therefore has n+1 producers. † This treatment is different from the one tested in Brandts et al. c1 [ q ] = ( n + 1) ⋅ cn +1 [((1 n + 1)) ⋅ q ] . The numbers we obtained after this rounding procedure were also the numbers we used to calculate the theoretical predictions. not by entry.12%. 15 and JPM for Joint Profit 12 If all the assets of n producers would be merged into one single firm. 13 To not unduly add to our subjects’ cognitive load. The total costs of the single firm would thus be c1 [ q ] = n ⋅ cn [ n ⋅ q ] . As the total assets base is held constant. 14 The prefix NE stands for Nash equilibrium. . Using (1) and c3 [ q] . 14 See Van Koten and Ortmann (2011). * These treatments are identical to the ones tested in Brandts et al. recall that the subjects can choose only integer production values. Table 2 contains the Nash equilibrium predictions for each treatment. p.10 for the precise rounding rules. and page 9. as the joint consumer and producer surplus realized in the experiment divided by the maximum joint consumer and producer surplus (the Walrasian level of joint surplus). (2008). As a result these rounding rules. They were motivated by the more concentrated markets of the new EU member states as well as related treatments in Le Coq and Orzen (2006). More on this below in section 4 (results section). for more details about the numerical procedure to determine the Nash equilibria. 15 We define efficiency. Likewise. we presented costs that were rounded. The cost function is transformed into12 cn +1 [q ] = ( n ( n + 1)) ⋅ cn [((n + 1) ( n )) ⋅ q ] .13 (1) Table 1: Treatments Without Forward Market With Forward Market 2 producers M2# M2F# 3 producers M3* M3F* 4 producers M4† – # These treatments were not tested in Brandts et al. the producers’ cost functions in M2 and M4 can be derived as c2 [q ] = (2 27)q 3 + (2 3) q 2 + (1 3) q and c4 [q ] = (22 27)q 3 + (4 3)q 2 + (1 2) q . (2008). respectively. who each have n ( n + 1) of the producers’ asset base in the market with n producers. Table 2 for an overview of the aggregate cost levels for M2(F). Walras for the efficient solution (the outcome that maximizes the total surplus). following Brandts et al. See Van Koten and Ortmann (2011). the given cost function of producers in M3. then this single firm would minimize its costs by 1 dividing production equally over the n plants. On average the absolute discrepancy is 0. c1 [ q ] is identical in both equations and thus the above formula can easily be derived. page 11. (2008).

9 NE M3 NE M3F NE M4 Walras (M2. (2008). and Walras (M4) in this experiment are identically to those in Brandts et al. M3F) . NE (M3). Explanation of row terms: qtif stands for the forward production of each producer.8 – 14/1518 43 839 29537 24381 53918 97. Walras (M3. defined as the aggregate production costs divided by the minimal aggregate production costs. for the particular parameterizations chosen. footnote 22 for on overview of the main (but inconsequential) differences. qt for the total aggregate production of forward (if it is present) and spot market for all producers together. and our predictions are almost identical to the ones reported in their paper. M2F) – 25/2619 51 623 21053 34425 55478 100 Walras (M3. the qualitative effect of introducing a forward market and of adding one or more competitor by symmetric divestiture on The market outcomes for JPM (M3. 19 One generator produces 26 units. since the level of production is predicted to be below the Walrasian level due to market concentration.(%) – 20 40 920 31520 21060 52580 94. Total. Namely. Table 2: Theoretical predictions electricity markets NE M2 NE M2F (two Nashequilibria) 2 20 40 920 31520 21060 52580 94. M3F). the other two 25 units. Welfare can also be affected by productive efficiency. Walras for the Walrasian (welfare optimizing) equilibrium. 18 One generator produces 15 units.6/19 Maximization (the monopoly solution). the other two 14 units.7 JPM (M3) JPM (M4) qtif qti qt pt Prod. M3F) – 17 51 623 21063 34425 55488 100 Walras (M4) JPM (M2. 16 .S for producer surplus.2 – 8 32 1136 33572 13392 46964 84.(%) for efficiency in percentage.9 – 13 52 596 19672 35802 55474 100 – 11 33 1109 33567 14256 47823 86. The theoretical predictions show.S for total surplus.2 5 15 45 785 27885 26730 54615 98. p t for the price. M2F) – 16 32 1136 33572 13392 46964 84.S Total S Welf. See Van Koten and Ortmann (2011).8 11 22 44 812 28768 25542 54310 97. Prod.7 Explanation of column terms: NE stands for the Nash-equilibrium.4 – 11 44 812 28768 25542 54310 97. qti for the total production of forward (if it is present) and spot market for each producer. Welfare is strongly affected by the level of aggregate production.17 We expect that the effect of productive efficiency is minimal and roughly equal across our treatments. any measure that increases aggregate production will most likely increase welfare. 17 Due to the convexity of the cost functions. JPM (M4). and JPM for joint profit maximization (collusion).S for consumer surplus.S Cons. Cons. Welf. minimal production costs are attained when production is distributed as evenly as possible across producers. NE (M3F).16 Following standard convention in experimental economics. we define the welfare percentage as the realized joint consumer and producer surplus in the experiment divided by the maximum joint consumer and producer surplus (the Walrasian level of joint surplus).

footnote 25. but if both Nash equilibria are chosen with some positive probability. Similarly. December 2009. Hypothesis 3 compares the predicted effect of the forward market relative to the benchmark. introducing a forward market will have a positive effect on quantity.7/19 aggregate production and welfare. . both located in Prague. As mentioned above. by the principle of insufficient reason. theory predicts that the forward market will have a positive effect on quantity. Since the production in M3 is in between the low and high Nash equilibria of M2F. The subjects earned on average 382 Czech Koruna per hour. we hypothesize that the effects will be equal.. we formulate hypotheses in Table 3Error! Reference source not found. The sessions with (without) a forward market lasted approximately 2 hours (90 minutes). for the details of the treatments. in M3. Using q(x) to denote aggregate production in market structure x. We have no prior. page 13. ) 3. including a show-up fee of 100 Korunas.20 Subjects were students at the Charles University or at the University of Economics. and April 2010 in Prague at CERGE-EI. Subjects thus earned on 20 See Van Koten and Ortmann (2011). Table 3: Hypotheses H1 H2 H3 Effect of forward market Effect of adding one competitor by symmetric divestiture (benchmark) Effect of forward market relative to benchmark q(M2F) > q(M2) q(M3F) > q(M3) q(M3) > q(M2) q(M4) > q(M3) q(M2F) = q(M3) q(M3F) > q(M4) Hypothesis 1 predicts the effect of introducing a forward market on aggregate production. Both measures are hypothesized to increase aggregate production in markets M2 and M3. Introducing a forward market in M2 results in two welfare-rankable Nash equilibria: one with quantity 40 (“low”) and one with quantity 44 (“high”). Experimental procedures The experimental sessions were conducted in October 2009. A total of 198 students participated in the experiment. theory predicts that the effect of the forward market will be larger than the benchmark. we cannot make a prediction. Hypothesis 2 predicts the effect of adding one more competitor by symmetric divestiture (the benchmark) on aggregate production. we expect the data on welfare to closely follow the patterns of those on the aggregate quantity. In M3. Therefore.

for the consolidated instructions. the resulting price per unit on the (forward) spot market.23 Producers were shown this forward-market demand schedule in the first period of each round. Each treatment consisted of 24 rounds. and subjects enacted the roles of producers and sellers. page 14. but were given an earnings table on the screen and as a printout. Our experiment was therefore well incentivized on the margin. minus the production cost. producers chose how many units to produce and sell in the forward market. 2007). The earnings were thus salient and in line with standard remuneration practices in experimental economics. marginal cost.e. and their own profit. The price per unit was determined by substituting the number of units sold by all producers in the forward and spot market together for Q in the demand schedule p[Q ] = Max[0. minus the production cost. The demand schedule was pre-programmed. 22 Le Coq and Orzen (2006) also employed pre-programmed traders. for details on the preprogrammed traders. subject could. These units were produced and delivered to traders in the second period (in the spot market). costs. The minimum earning was 330 and the maximum earning was 1080 Korunas. At the end of the period. This amount equaled about €26 (and about €36at official purchasing power parity.22 Traders were programmed to act rationally. 1981). The preprogrammed traders sold all the units they bought. During the experiment. producers were paid the number of units they produced in the forward market times the price per unit. (2008) use experimental subjects as traders and find that traders earn only a small fraction (about 8%) of the profits that a producer earns. 21 . calculate the competitor’s contribution. In the first period. however. Earlier experimental evidence indicates that the presence of strategically acting middlemen generally does not alter allocations and that the profit of middlemen converges to zero quickly (Plott and Uhl. and cumulative costs. which we presented to our producers. and even more at a student-specific purchasing power parity). Appendix F. 23 See Van Koten and Ortmann (2011). which they could use to inform the number of units to offer in the forward market. two pre-programmed traders competed in prices for the total number of units that were offered. 2000 − 27Q ] . In the forward market. The market simulation was programmed in Z-tree (Fischbacher. producers chose the number of units to produce and sell in the spot market. M2 and M2F). In treatments with two subjects (i. Subjects were not shown their competitors’ profits. All producers were paid the number of units they produced in the spot market times the price per unit. they were shown the following outcomes: the aggregate quantity sold. The manner in which traders are represented in the experiment should not significantly affect outcomes. as traders are middlemen (between producers in the forward market and end demand in the spot market). They were not shown the demand schedule. Brandts et al.8/19 average 640 Korunas21. In the second period of each round. Treatments with a forward market contained two periods per round: the first for the forward market and the second for the spot market. The same experimenter read the (English language) instructions to the subjects for all sessions. or contributions to the aggregate quantity sold. once subjects submitted their choice for the (forward) spot market. Their actions defined a demand ) schedule. and page 45.

25 ) Figure 1 shows the evolution of total (aggregate) quantities sold per period. Indeed subjects need several rounds of trading to become familiar with the laboratory environment and before they react to the embedded incentives. as did Le Coq and Orzen (2006). with three producers by triangles. M2F. M4 Including all 24 rounds. (2008) did. (2008). Results For our statistical tests. averaged over groups in each treatment. M3 b) M3. Wilcoxon rank-sum tests (rank-sum tests). unless indicated otherwise. Production stabilizes between rounds 8 and 12. or only the last 10 rounds as Brandts et al. Since each participant took part in one experimental session only.24 Following Le Coq and Orzen (2006). we test for deviations from the Nash-equilibrium predictions using two-sided Wilcoxon one-sample signed-rank tests (two-sided signed-rank tests). These tests confirmed most of the results presented here. The error terms are adjusted for clustering at the group level by using the robust Huber-White sandwich estimator (Froot. M3F. and with four producers by squares. p. M3F. Treatments with two producers are represented by circles. Figure 1: Aggregate production a) M2. M3. such as in relatively easy auction experiments (Hertwig and Ortmann. we treat each set of sellers (“a group”) per treatment as a single data point. In line with Brandts et al. The trade volume in all treatments is initially low26. To be conservative. M2. Loss-averse behavior due to the initially considerable uncertainty could be another explanation. The results of these tests can be found in Van Koten and Ortmann (2011). We therefore have 11 statistically independent data points for each of the five treatments reported here. The treatments without forward markets are represented by open circles.1. 26 It is likely that these trajectories are anchored by the examples in the instructions in which we used low numbers to facilitate understanding of the basic relationships. and M4. we use the last 12 (out of a total of 24) rounds of the data because the experiment is complicated. There was no bankruptcy in the experiment.31. the treatments with forward markets by filled circles or triangles. Appendix B. data points are independent across treatments. but then quickly converges to the Nashequilibrium. triangles or squares. M2F. without a constant (F-tests): AggregateSupply = β1 ⋅ M 2 + β 2 ⋅ M 2 F + β 3 ⋅ M 3 + β 4 ⋅ M 3F + β5 ⋅ M 4 + ε . 24 . does not change our results qualitatively and all results remain significant. we ran robustness tests using. 25 In addition. 2001). we test our hypotheses with F-tests based on an OLS regression of the dependent variable on the 5 treatment dummies.9/19 4. 1989).

2 (4. but not significantly different from the high Nash-equilibrium (p-value = 0.010) 45 110.32). (2008) for comparison The first number gives the percentage of output relative to the low production Nash-equilibrium.6 (2.2 (3. production in M2F is significantly higher than the low Nash-equilibrium (p-value = 0. Le Coq and Orzen. and Oechssler. 2004). the M2 and M3 quantities are not significantly different from the Nash-equilibrium predictions (two-sided signed rank test.g. 27 .816) 40/44 116 %/ 105%27 N = 11 M3 44.0% N = 11 Results of Brandts et al.3 (5. regressing aggregate production on the period of the experiment shows a significant upward slope. We see no evidence for stable collusion. For the treatments with a forward market. 2006. which suggests that over time and as subjects become more experienced with the task.063) 43 102. In line with earlier findings (e.004). they become less likely to collude. Also. production in M3F is significantly higher than the Nash-equilibrium (p-value = 0. whereas the M4 quantities are significantly larger (p-value = 0.033) 40 98. ) when the number of competitors is equal to two (three or four) and there is no forward market.10/19 55 Aggregate Production 35 40 45 50 30 1 4 8 12 Period M2 M3 16 M2F 20 24 30 1 Aggregate Production 35 40 45 50 55 4 8 12 Period M3 M4 16 M3F 20 24 For the treatments without forward markets. the second number relative to the high production Nash-equilibrium.068). Table 4 Average aggregate production (last 12 rounds) Average aggregate production Predicted quantity (NE) % of NE prediction Number of observations M2 39. production tends to be smaller (larger) than the Nash-equilibrium.26) 44 105.248).021).7% N = 11 M2F 46.3 (6. indeed the data suggest the opposite. Huck.9% N = 11 M3F 49. both p-values > 0.1% N = 11 M4 46. Namely. Normann.

9.0% versus 105. A detailed analysis can be found in Van Koten ) The standard error is computed based on the averages for each of the 11 independent groups over the last 12 rounds. Admittedly. Our results replicate the findings of Brandts et al.5 48. The use of pre-programmed traders reduces the variation in subject choice.06). 2003. 89).11/19 Average aggregate 42.8%.02). unhindered effect. and this may enable the forward market to exert a stronger. and found that the average aggregate production was significantly lower than in treatments with (convex) production costs. Number of N = 15 N = 15 observations 50.86) 49 105. p-value = 0.9%. (2008) (46. 56) and M4 (105. as Wilcoxon rank-sum tests show no significant differences in “% of NE prediction” across studies for M3 (102. average aggregate production and % of NE prediction of Brandts et al. this argument is speculative. This conforms to earlier theoretical and experimental evidence. 93. we observed larger average aggregate production in markets with two producers than in Le Coq and Orzen (2006). suggesting a behavioral effect that further strengthens the competitive effect of a forward market. pvalue = 0. which shows that increasingly steep cost curves lead to more competitive outcomes (Davis & Reilly.0% N=7 113. 29 28 .6%.. p-value = 0.9% 103.6% Number of N=7 N=6 observations Results of Le Coq and Orzen (2006) for comparison % of NE prediction 93.29 Using the percentage of the Nash equilibrium predictions measure. (2008): outcomes are close to the Nash equilibrium predictions for M3 and significantly more competitive for M4.57) (2. We find.28 Row “% of NE prediction” shows average aggregate production per treatment group relative to the Nash-equilibrium prediction (which is given in the row above).0%. which suggests the asset effect influenced their results. however.2 versus 50.1% versus 103.41) production Predicted quantity 43 45 (NE) % of NE prediction 98. (2008) and Le Coq and Orzen (2006) are also shown in Table 4. (110. (2008).9% versus 98.8 (5. (2008) for the treatments without forward markets. Average aggregate production in the market with four producers (M4) is significantly smaller than in Brandts et al. pvalue = 0. 2%. we conducted treatments without production costs. Engel 2007). For comparison purposes.7% N=8 The shaded columns indicate the treatments that are identical to the ones tested in Brandts et al. Our use of pre-programmed traders may be the cause of our stronger effect of the forward market compared with Brandts et al. Table 4 shows the average aggregate production for the last 12 rounds per treatment group with the standard deviation in parentheses. As robustness tests. a significantly higher average aggregate production in M3F than Brandts et al. We thus find the same behavioral effects as Brandts et al.9 (5.

2 -> 46.003) ***(p<0. we also compared the averages for the groups using a two-sample Wilcoxon rank-sum (MannWhitney) test. 31 Also in our robustness sessions for M2 and M2F.q(M3F) ≤ q(M3) is REJECTED in favor of q(M3F) > q(M3) . p-value = 0. We find therefore find partial support for Hypothesis 2: .105. The increases are strongly significant and are qualitatively in line with all previous experimental evidence. .001.30 Table 5: Effects of one more competitor and forward market H1 H2 H3 Effect of forward market Effect of adding one competitor by symmetric divestiture (benchmark) Effect forward market relative to benchmark a) q(M2F) > q(M2) b) q(M3F) > q(M3) a) q(M3) > q(M2) b) q(M4) > q(M3) a) q(M2F) = q(M3) b) q(M3F) > q(M4) ***(p=0. In markets with 3 producers. Appendix B. c (which becomes significant). introducing a forward market increases aggregate production by 12% (44.12/19 and Ortmann (2011. the session with a forward market has significantly higher aggregate production that the one without a forward market (p<0. but there misses statistical significance at conventional levels albeit barely.b (which becomes insignificant) and Hypothesis 3. p-value = 0. as predicted. where producers had zero costs.of adding one competitor by symmetric divestiture): In markets with 2 producers.105) (p=0.3 -> 46. the benchmark measure of adding one more competitor by divestiture increases.001) ** (p=0.2). Table 5 presents the results of the F-tests based on OLS regressions. The hypotheses accepted (rejected) are the same. except for Hypothesis 2.010). See the Appendix for a detailed analysis.2) in markets with two competitors and by 5% (44. In line with theoretical predictions. adding one more competitor by divestiture increases production by 12% (39. .q(M3) ≤ q(M2) is REJECTED in favor of q(M3) > q(M2).q(M4) ≤ q(M3) is NOT REJECTED in favor of q(M4) > q(M3) .q(M2F) ≤ q(M2) is REJECTED in favor of q(M2F) > q(M2).3) in markets with two ) competitors. Results testing Hypothesis 2 (Effect of benchmark .31 .006) (p=0.2) in markets with three 30 To test robustness.374) ***(p=0. clustered on groups.6) in markets with three competitors and by 18% (39. production significantly.3 -> 44.002) N=528 (11) N=792 (11) N=660 (11) N=924 (11) N=660 (11) N=924 (11) N: Number of observations (independent groups) Results testing Hypothesis 1 (Effect of forward market): Introducing a forward market increases production. In line with theoretical predictions. p-value = 0. p-value < 0.006.2 -> 49. it also increases production. We therefore find support for Hypothesis 1: .003.

6). The increase is strongly significant in markets with two producers. the latter as predicted. while a market with two producers is relatively collusive (producing slightly less than the Nash equilibrium quantity). The result is in line with the finding of Huck et al.5% + 5%# + 20.9% *** + 12. p-value = 0.13/19 competitors. which is not significantly different from zero (p < 0.32. the effect of productive inefficiency turns out to be small and inconsequential. Table 6 Comparison of our results with those of earlier studies Theoretical predictions in our study Results of earlier studies Empirical results of our study One more competitor (OMC) Forward Market 32 Market with 2 competitors + 7. p-value = 0.001 (0. but not in markets with 2 competitors. We therefore find partial support for Hypothesis 3: . (2004) that a market with three producers is already relatively competitive (producing slightly more than the Nash equilibrium quantity).93).2) + 17. as discussed earlier. As expected.2 -> 46.3) and 7% in markets with three competitors (46.01) over the last 12 rounds.8% *** See Van Koten and Ortmann (2011) for details on the effect of our treatments on the welfare and productive inefficiency measures. The difference is 5% in markets with two competitors (44.q(M2F) = q(M3) is NOT REJECTED in favor of q(M2F) ≠ q(M3). The difference is statistically significant only in the market with three competitors. The declining effects of larger numbers of competitors is of interest due to the typical market configuration in old and new member states. together with the key results of earlier experiments. .q(M3F) ≤ q(M4) is REJECTED in favor of q(M3F) > q(M4) .2 -> 49. . Table 6 summarizes our theoretical and experimental results for the aggregate production.002.1% ** (39. Running the same tests for welfare confirms our) expectation that the welfare percentage closely follows the patterns of aggregate production. Results testing Hypothesis 3: Introducing a forward market increases production more than the benchmark in markets with 3 competitors.374.3 -> 44. We also observed that prices in the forward market were rational in the sense that they predicted prices in the spot market well: the average difference between the spot and forward market was equal to 0. Introducing a forward market increases aggregate production more than adding one more competitor by divestiture.

33 . the competition-enhancing effect of introducing an additional competitor to the market is minor in a market with three competitors. Kujal and Rassenti (2009) that started circulating while we were writing our paper.33 Introducing a forward market increases aggregate production significantly more than adding one more competitor. by 2% (high NE) (Le Coq & Orzen. by 9. 2006) - (39. Our results show that in markets with three competitors. by 8% (low NE) FM.5%** (Brandts et al. Appendix D.2 -> 49. Our findings show that without an asset effect. adding one more competitor increases aggregate production.5% (44.3% + 19. introducing a forward market significantly increases aggregate production. 2008). by 2. In markets with two competitors. significant effect in Brandts et al.. our experienced subjects supply a slightly higher production which is in line with the experimental literature on the effect of experience on public good provision (Ledyard. (2008) that this increase would be smaller than that of adding one more competitor. Our data suggests the opposite. 2008 ) + 9..2) (not significant) + 4. p.. Ferreira et al. which is in line with our theoretical prediction and earlier experimental results (Brandts et al.3 -> 46. 2008) + 4.3% *** (46.3) FM. by 4. 2006). introducing a forward market significantly increases aggregate production.7% (44. in line with earlier experimental results (Le Coq and Orzen.2% ** (Brandts et al.6%*** (Brandts et al.6) FM. see our Van Koten and Ortmann (2011). but contradicts the findings of Brandts et al. 2008 ) OMC. (2009) find that forward markets have a positive effect on the aggregate production with inexperienced subjects. The increase is. which – as we have argued – makes production relatively cheaper. 39. if the competitive-enhancing effect is robust when subjects are experienced.14/19 (FM) Largest increase by OMC. For the detailed results. however. but no – or a negative – effect with experienced subjects. not significant. We tested. We became aware of the possibility of such an issue through a working paper of Ferreira. as a robustness test.7% + 12.3) (not significant) Market with 3 competitors One more competitor (OMC) Forward Market (FM) Largest increase by + 2. We do not find that experienced subjects have a lower aggregate production than inexperienced subjects.3% Results contrast with earlier results.6) FM.0% *** (44. by 7. which is in contrast with the findings of Brandts et al.2 -> 46. We could not replicate the finding of Brandts et al. Results contradict earlier results. The larger. (2008) is probably due to the asset effect. In line with our theoretical prediction.2 -> 49. On the contrary. which confirms our theoretical prediction.2 -> 46. (2008) (the contradictory findings are indicated by a bold frame in Table 6). 1995). (2008)..

15/19 6. Their result seems driven in part by confounding the competition effect and the asset effect. The result that introducing a forward market has a competition-enhancing effect is in line with the empirical studies of Wolak (2001) and Van Eijkel and Moraga-Gonzalez (2010). Such a policy might improve on the effectiveness of forward markets in the EU.127).50). who found a stronger effect for the structural remedy of adding one more competitor than for the behavioral remedy of introducing a forward market. This gives tentative support to the European Commission’s preference for behavioral remedies. Our results suggest the behavioral remedy of introducing a forward market in concentrated markets with two or three competitors is an effective remedy for countering market power by increasing the aggregate production. we find that the effect of the structural remedy of adding one more competitor is weaker and is now dominated by the effect of the behavioral remedy of introducing a forward market. In addition. p.reducing concentration by adding an additional competitor by divestiture. Conclusion We experimentally investigated the effects of introducing a forward market on competition. (2008). This asset effect is likely influential. who found indications that forward trading increased aggregate supply in the Australian electricity market and the Dutch gas market. 2007a. such as the introduction of a forward market (European Commission. This reduces the怀 aggregate cost and thus increases the incentive to raise production. quarters. we find a stronger behavioral effect of introducing a forward market in a market with three producers. In our study. as they do not incorporate the costs of the increase in the asset base (the cost of building extra production plants). as producers have steeply increasing costs. in France the PowerNext exchange market allows for the trading of forward and future contracts of months. the EU has no single policy towards the design of forward markets for electricity. (2008) add one more competitor by entry. which increases competition and increases the aggregate asset base. In Spain. p. p.127). In contrast.11). 2007b. p. respectively. Our results contradict the findings of Brandts et al. for example. as it has made trading in the pool mandatory (European Commission. In Greece forward trading has been made virtually impossible by design. we control for the asset effect by adding one more competitor by divestiture. Our study suggests that forward markets could play a role in enhancing competition in concentrated markets and thus suggests that the design. of public forward exchanges such . The welfare effects these authors report are not conclusive. or evolution. As a result of these two differences. 2007a. At present. forward trading is de facto forbidden by design (European Commission. We compared the effect to the best alternative . as design is an important factor for the thickness of forward markets in EU countries (European Commission. Brandts et al. 2006a. and years ahead.

. and Vila. research grant P402/11/0364 from the Grant Agency of the Czech Republic (GACR) and the Loyola de Palacio Chair at the RSCAS of the European University Institute. Werner Güth. Anna Gunnthorsdottir. 1376–1405. The Economic Journal 118 (January). the Humboldt University. however. Journal of Economic Theory 59(1). and Schram. 1997. 192–214. the European Energy Market conference 2012. research center grant No. F. Aleksandar Zaklan. and the Max-Planck Institute for Economic Research in Jena. 2010. Bushnell. J. Energy markets are. 9-11). J. References Allaz. Measuring market inefficiencies in California’s restructured wholesale electricity market. B. y We are grateful for financial support from the REFGOV Integrated Project which was funded by the 6th European Research Framework Programme . the public observability of forward positions is essential for the competition-increasing effect of Allaz and Vila (1993) to arise (Hughes and Kao. and Wolak. the Australian School of Business. P. S. Competition with forward contracts: a laboratory analysis motivated by electricity market design. Hannes Weigt.16/19 as in France (and many other developed markets) should indeed be encouraged. Moreover. J. Forward trading positions may thus often not be observable. UNSW. 2002. participants at the ESA 2010 conference. LC542 of the Ministry of Education of the Czech Republic implemented at CERGE-EI. two anonymous referees of the EER.. the University of Sienna. Borenstein.. Axel Ockenfels. 1–16. Pezanis-Christou. the European University Institute. for their excellent comments. Cournot competition. and participants of seminars at CERGE-EI. Our results thus suggest that making the EU forward energy markets more transparent could contribute to a more competitive market. Paul Pezanis-Christou. not completely transparent as most trading takes place at Over-The-Counter (OTC) markets over different trading platforms and a considerable proportion (around 10%) takes places bilaterally without intermediaries (European Commission. Hodaka Morita. Arthur Schram.CIT3-513420. forward markets and efficiency. Christian Redl. Gregory Swinand. 2010). Van Eijkel and Moraga-Gonzalez. 2008.-L. Acknowledgements We thank Libor Dušek. Brandts. American Economic Review 92(5). A. 1993.

DG Competition report on energy sector inquiry. H. A. Departamento de Economía. Le Coq.. L. R.M. S. 1996. 2001. K. D. D. Kennedy School of Government. U.. SEC(2010) 1511 final). 1997. Experimental Economics 6. Kamat. Huck. 2011. 1992. Universidad Carlos III. Strategic forward contracting and observability.. H. 491–549. 171-178. R. Z-Tree: Zurich Toolbox for Ready-made Economic Experiments.17/19 Davis. Accompanying document to the Proposal for a regulation of the European parliament and of the council on energy market integrity and transparency. California Electricity Prices and Forward Market Hedging. 209-226. Green. J. S. B. S.W. S. 415–431. Duso. and Oechssler. Journal of Economic Behavior and Organization 53(4). J. 121-133. Competition in the British electricity spot market.. 383–451. Journal of Political Economy 100(5). The Journal of Industrial y economics 44(2). J. 2007. L.S. 2007. Economics Working Papers we092616. European Commission 2007b. 2006.. P. 2004. How much collusion? A meta-analysis of oligopoly experiments. R. Journal of Economic Behavior and Organization 61(3). The strategic motive to sell forward: experimental evidence. 5–37. Utilities Policy 14. -T. Market power mitigation in the UK power market. 2003. W. 205-216.B. Oren. Prospects for the internal gas and electricity market. Two are few and four are many: number effects in experimental oligopoly. R. 2000. Kujal. and Rassenti. 435–446. Harvey. 2009. Experimental practices in economics: A methodological challenge for psychologists? Behavioral and Brain Sciences 24. Green. (COM(2010) 726 final. Normann. How effective is European merger control? European Economic Review 55. R. Increasing competition in the British electricity spot market.. Do forward markets enhance competition? Experimental evidence... John F. Two-settlement systems for electricity markets under network uncertainty and market power. R. Cost structures and Nash play in repeated Cournot games. 980–1006. C. C.. and Yurtoglu. Mimeo. S. 2006. and Newbery. J. T. 929–953. Journal of Competition Law and Economics 3(4). and Orzen. 2004. Fischbacher. Engel. Hogan. Harvard University. Hertwig. Experimental Economics 10(2). European Commission 2007a. and Ortmann. European Comission 2010. Ferreira. Hughes. and Reilly. D. Green. and Kao. J. International Journal of Industrial Organization 16. Journal of Regulatory Economics 25. Gugler. . 76-89.

In: Kagel. 515–544. Southern Economic Journal 47(4). O. . Wolak. IESE Business School Working Paper No. 2001. 1995. Plott. The Economist as Engineer: Game Theory. Public goods: a survey of experimental research. 1-39. 2006. Problems of liberalising the electricity industry. T.ssrn. Competitive Equilibrium with Middlemen: An Empirical Study. Princeton University Press. 3128–3140.. Van Eijkel. Alvin E. Rassenti. pp. B. The unbundling regime for electricity utilities in the EU: A case of legislative and regulatory capture?. A. J. Handbook of Experimental Economics. Roth.L. Van Koten.bertwillems. V. The Cato Journal 21(3). & Roth. 2011. B. and Moraga-Gonzalez. R. J. Virtual divestitures. Princeton. 2002.cfm?abstract_id=1835482 Willems. will they make a difference? Available at www. 1063-1071. Newbery. Econometrica 70. Structural Versus Behavioral Remedies in the Deregulation of Electricity Markets: An Experimental Investigation Guided by Theory and Policy Concerns. A. Do firms sell forward for strategic reasons? An application to the wholesale market for natural gas. Experimentation. y Available at http://papers.). S. 1341-1378. J. 1981. 2002. S. and Ortmann. R. 864. H.18/19 Ledyard. Using experiments to inform the privatization/deregulation movement in electricity. Van Koten. European Economic Review 46(4– 5). C.com. 2010. (Eds. Energy Economics 30(6). E. 919–927. and Wilson. F. D. S. and Computation as Tools for Design Economics. 2008. An empirical analysis of the impact of hedge contracts on bidding behavior in a competitive electricity market. International Economic Journal 14. A.com/sol3/papers. 2002. 111–194. A. and Uhl. and Ortmann. Smith.

Likewise. c1 [ q ] is identical in both equations and thus the above formula follows.34 With convex costs. assuming convex costs. ′ The first order condition for profit maximizing is p[Q ] − cn [qi ] = qi ⋅ − p ′[Q ] . Focusing on symmetrical equilibria for producers with identical cost functions. but less so than in the case of entry. the divestment causes the marginal cost to increase: ′ ′ ′ cn +1 [q ] = cn [(( n + 1) n ) ⋅ q] > cn [ q] . ′′ where cn [ q] ≥ 0 and P ′[Q ] < 0 . c1 [ q ] = ( n + 1) ⋅ cn +1 [((1 n + 1)) ⋅ q ] . as y argued above. The competition-enhancing effect of adding more competitors will thus be lower with divestiture than with entry. is equal to zero. As Le Coq and Orzen (2006) assume zero production costs. Adding competitors by entry increases n and thus lowers the right hand side of (3). However. each of the n producer maximizes: π i [qi ] = qi ⋅ p[Q ] − cn [qi ] . this can be rewritten as ′ p[Q ] = (Q n ) ⋅ − p ′[Q ] + cn [Q n ] . the left hand side must decrease. increases. (2008). to preserve equivalence. There is no such difference in effect between entry and divestiture when marginal costs are ′ constant as this implies that the marginal cost term in (3). Q. the cost function is transformed into cn +1 [q ] = ( n ( n + 1)) ⋅ cn [((n + 1) ( n )) ⋅ q ] . 34 If all the assets of n producers would be merged into one single firm. as (2) indicates that the marginal cost function of each producer. their benchmark of adding competitors can be interpreted as the result of either entrance or of entry. outlays that are (3) not necessary for the introduction of a forward market. Thus. (2) (1) In a market with Cournot competition. which implies that the aggregate quantity Q increases. is thus smaller. As the total assets base is held constant. c′ [⋅] . then this single firm would minimize its costs by 1 dividing production equally over the n plants. cn [Q n ] . Adding competitors by divestment increases n and lowers the right hand side of (3).19/19 Appendix When one competitor is added to a market with n producers by symmetric divestiture. The total costs of the single firm would thus be c1 [ q ] = n ⋅ cn [ n ⋅ q ] . For the setup of Brandts et al. entry and divestiture have differential effects. n The increase in the aggregate production. . The entry of new competitors therefore increases the aggregate production. needed to keep the equivalence of (3). this involves also a costly investment in production assets by the entrant.

Sign up to vote on this title
UsefulNot useful